SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2006.04902 · 2020

What Matters in Unsupervised Optical Flow

arXiv · PDF · Open in the Atlas

Code that ran

16 of 17 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (0 of them constructed an object rather than computing a result; 1 honoured the contract we drafted, 0 violated it and 15 have no contract verdict; 0 failures of our instrument; 1 no recorded run), measured 2026-11-22; repository links from the Papers with Code archive attributed under CC BY-SA, no affiliation; bodies inline only under a permissive licence as GitHub detects it. Not a reproduction of the paper's results.

"Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results. This record, dated, as JSON: /api/ran/2006.04902.json · badge: /api/ran/2006.04902.svg

Each repository's role is a label: a classifier's output recorded on the link, or, where no role is recorded, what the Papers with Code archive marked. The line under it names its basis. Neither says who wrote the code.

RepositoryRoleRan
junbongjang/contour-tracking community (classifier label)
classifier basis not recorded on the link
16 ran · 1 no recorded run
FunctionStatusWhere it lives
compute_cost_volume Ran
ran on a synthesized input
junbongjang/contour-tracking/src/contour_flow_model.py
code served (permissive licence) · get_code("f1a6a5a446ff10ba")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned EagerTensor, ndim 4, dtype family float, shape (2, 8, 6, 25) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import tensorflow as tf

# Set seed for reproducibility
tf.random.set_seed(0)

# Create small test tensors
batch_size = 2
height = 8
width = 6
channels = 4
max_displacement = 2

# Create features1: [batch, height, width, channels]
features1 = tf.random.normal([batch_size, height, width, channels])

# Create features2: [batch, height, width, channels]
features2 = tf.random.normal([batch_size, height, width, channels])

ARGS = {
    "features1": features1,
    "features2": features2,
    "max_displacement": max_displacement,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
tensorflow — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type EagerTensor, ndim 4, dtype family float, shape (2, 8, 6, 25) — shape (2, 8, 6, 25) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
corr_2d_loss Ran
ran on a synthesized input
junbongjang/contour-tracking/src/tracking_utils.py
code served (permissive licence) · get_code("e3a04fdb34e68bb4")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned EagerTensor, ndim 0, dtype family float, shape () as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import tensorflow as tf

# Set random seed for reproducibility
tf.random.set_seed(0)

# Create small tensors with distinct dimensions
batch_size = 2
num_tracked_points = 3
num_seg_points = 5
num_classes = 4

# gt_prev_id_assignments: shape (B, num_tracked_points, 1), dtype int32
# Contains indices into the num_seg_points dimension
gt_prev_id_assignments = tf.constant(
    [[[1], [2], [0]], 
     [[3], [1], [4]]], 
    dtype=tf.int32
)

# gt_cur_id_assignments: shape (B, num_tracked_points, 1), dtype int32
# Contains class labels for sparse categorical crossentropy
gt_cur_id_assignments = tf.constant(
    [[[0], [1], [2]], 
     [[3], [0], [1]]], 
    dtype=tf.int32
)

# corr_2d_matrix: shape (B, num_seg_points, num_classes), dtype float32
# This will be sampled by index to get logits for the loss function
corr_2d_matrix = tf.random.normal(
    shape=(batch_size, num_seg_points, num_classes),
    dtype=tf.float32
)

ARGS = {
    "gt_prev_id_assignments": gt_prev_id_assignments,
    "gt_cur_id_assignments": gt_cur_id_assignments,
    "corr_2d_matrix": corr_2d_matrix,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
tensorflow — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type EagerTensor, ndim 0, dtype family float, shape () — shape () as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
get_closest_indices_from_gt_and_pred Ran
ran on a synthesized input
junbongjang/contour-tracking/src/preprocessing/contour_tracking_manuscript_figures.py
code served (permissive licence) · get_code("bc7e65890c171b15")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned list, ndim 1, dtype family int; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import numpy as np

# Create sample ground truth points
# Shape: (2, 4) - 2 ground truth points, each with 4 dimensions
init_gt_points = np.array([
    [0.0, 1.0, 2.0, 3.0],
    [1.5, 2.5, 3.5, 4.5]
], dtype=np.float32)

# Create sample predicted points
# Shape: (3, 4) - 3 predicted points, each with 4 dimensions
pred_points = np.array([
    [0.1, 1.1, 2.1, 3.1],
    [2.0, 3.0, 4.0, 5.0],
    [1.4, 2.4, 3.4, 4.4]
], dtype=np.float32)

ARGS = {
    "init_gt_points": init_gt_points,
    "pred_points": pred_points
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type list, ndim 1, dtype family int — a list output has no array shape of its own; none is stated
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
get_image_name Ran
ran on a synthesized input
junbongjang/contour-tracking/src/preprocessing/MATLAB_tracking_points.py
code served (permissive licence) · get_code("70452baab57f53ae")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned str. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import numpy as np

def get_image_name(image_path, image_format):
    return image_path.split('\\')[-1].replace(image_format, '')

# Setup code
torch.manual_seed(0)

# The function takes:
# - image_path: a string that contains backslashes and ends with filename + format
# - image_format: a string representing the file format (e.g., '.jpg', '.png')

ARGS = {
    'image_path': r'C:\Users\Data\Images\photo.jpg',
    'image_format': '.jpg'
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type str — a str output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a str output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
get_ordered_contour_points_from_mask Ran
ran on a synthesized input
junbongjang/contour-tracking/src/preprocessing/main_process_tracking_points.py
code served (permissive licence) · get_code("89b3af1de1b9d022")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned ndarray, ndim 2, dtype family int, shape (84, 2) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import cv2
import numpy as np

# Create a small binary mask with a clear contour
# The function expects a mask where values are 0 or 1 (will be multiplied by 255)
# We'll create a simple mask with a circular region
a_mask = np.zeros((50, 50), dtype=np.uint8)
cv2.circle(a_mask, (25, 25), 15, 1, -1)  # Draw filled circle with value 1

ARGS = {
    "a_mask": a_mask
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
cv2, numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, ndim 2, dtype family int, shape (84, 2) — shape (84, 2) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
move_prev_tracked_point_and_align_to_smoothEdge Ran
ran on a synthesized input
junbongjang/contour-tracking/src/preprocessing/MATLAB_tracking_points.py
code served (permissive licence) · get_code("c6232c156aca7856")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned int, ndim 0, dtype family int, shape () as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import numpy as np

torch.manual_seed(0)

# Construct numpy arrays for the function parameters
# cur_smoothedEdge: array of 2D points
cur_smoothedEdge = np.array([
    [10.0, 20.0],
    [15.0, 25.0],
    [20.0, 30.0],
    [25.0, 35.0]
], dtype=np.float32)

# next_smoothedEdge: array of 2D points for alignment
next_smoothedEdge = np.array([
    [12.0, 22.0],
    [18.0, 28.0],
    [22.0, 32.0],
    [28.0, 38.0]
], dtype=np.float32)

# cur_protrusion: displacement vectors (same shape as cur_smoothedEdge)
cur_protrusion = np.array([
    [1.0, 2.0],
    [2.0, 1.5],
    [1.5, 2.5],
    [3.0, 1.0]
], dtype=np.float32)

# cur_smoothedEdge_index: integer index into cur_smoothedEdge
cur_smoothedEdge_index = 1

ARGS = {
    "cur_smoothedEdge": cur_smoothedEdge,
    "next_smoothedEdge": next_smoothedEdge,
    "cur_protrusion": cur_protrusion,
    "cur_smoothedEdge_index": cur_smoothedEdge_index
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type int, ndim 0, dtype family int, shape () — shape () as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
normalize_features Ran
ran on a synthesized input
junbongjang/contour-tracking/src/contour_flow_model.py
code served (permissive licence) · get_code("d0f7630cddfa3a3a")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned list, ndim 5, dtype family float; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import tensorflow as tf

# Create small feature tensors with shape [b, h, w, c]
# Using distinct dimensions: batch=2, height=3, width=4, channels=5
feature1 = tf.random.normal([2, 3, 4, 5])
feature2 = tf.random.normal([2, 3, 4, 5])
feature3 = tf.random.normal([2, 3, 4, 5])

ARGS = {
    'feature_list': [feature1, feature2, feature3],
    'normalize': True,
    'center': True,
    'moments_across_channels': False,
    'moments_across_images': False,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
tensorflow — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type list, ndim 5, dtype family float — a list output has no array shape of its own; none is stated
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
normalize_points Ran
ran on a synthesized input
junbongjang/contour-tracking/src/metrics.py
code served (permissive licence) · get_code("d654e53ca581da7f")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned ndarray, ndim 2, dtype family float, shape (2, 2) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import numpy as np

# Construct arguments for normalize_points
# The function expects:
# - pts: an array-like object with shape (..., 2) containing point coordinates
# - image_width: numeric value for image width
# - image_height: numeric value for image height

# Create a numpy array of points with shape (2, 2) - 2 points, each with (x, y)
pts = np.array([[100.0, 150.0], [200.0, 300.0]], dtype=np.float32)

# Image dimensions
image_width = 640
image_height = 480

ARGS = {
    "pts": pts,
    "image_width": image_width,
    "image_height": image_height,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, ndim 2, dtype family float, shape (2, 2) — shape (2, 2) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
point_wise_spatial_accuracy Ran
ran on a synthesized input
junbongjang/contour-tracking/src/metrics.py
code served (permissive licence) · get_code("5a5cd48d24333dfe")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned ndarray, ndim 1, dtype family int, shape (2,) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import numpy as np

# Create sample point sets
# ps1 and ps2 should be numpy arrays of shape (N, 2) representing (x, y) coordinates
ps1 = np.array([
    [100.5, 200.3],
    [150.2, 250.8]
], dtype=np.float32)

ps2 = np.array([
    [102.1, 202.5],
    [151.0, 249.2]
], dtype=np.float32)

# Image dimensions
image_width = 640
image_height = 480

# Threshold for spatial accuracy (in normalized coordinates)
thresh = 0.1

ARGS = {
    "ps1": ps1,
    "ps2": ps2,
    "image_width": image_width,
    "image_height": image_height,
    "thresh": thresh
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, ndim 1, dtype family int, shape (2,) — shape (2,) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
prevent_trajectory_cross Ran
ran on a synthesized input
junbongjang/contour-tracking/src/preprocessing/contour_tracking_manuscript_figures.py
code served (permissive licence) · get_code("5e16d4d82c56238b")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned ndarray, ndim 1, dtype family int, shape (5,) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import numpy as np

# Construct a valid argument for prevent_trajectory_cross
# The function expects pred_tracking_points_contour_indices to be an array-like
# that supports indexing, iteration, and np.amax()

pred_tracking_points_contour_indices = np.array([10, 5, 3, 8, 2], dtype=np.int32)

ARGS = {
    "pred_tracking_points_contour_indices": pred_tracking_points_contour_indices
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, ndim 1, dtype family int, shape (5,) — shape (5,) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
remove_points_touching_image_boundary Ran
ran on a synthesized input
junbongjang/contour-tracking/src/preprocessing/main_process_tracking_points.py
code served (permissive licence) · get_code("4c9c5b3ef68583d8")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned ndarray, ndim 2, dtype family int, shape (4, 2) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import numpy as np

# Create a sample image (height=10, width=8)
a_image = np.zeros((10, 8), dtype=np.uint8)

# Create ordered contour points as a 2D numpy array
# Points are in (x, y) format where x is column and y is row
# Include some points touching boundaries and some interior points
ordered_contour_points = np.array([
    [0, 5],      # touching left boundary (x=0)
    [2, 3],      # interior point
    [4, 1],      # touching top boundary (y=0)
    [5, 5],      # interior point
    [7, 6],      # touching right boundary (x=7, which is width-1)
    [3, 9],      # touching bottom boundary (y=9, which is height-1)
    [4, 4],      # interior point
], dtype=np.int32)

ARGS = {
    "ordered_contour_points": ordered_contour_points,
    "a_image": a_image,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, ndim 2, dtype family int, shape (4, 2) — shape (4, 2) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
reorder_contour_points Ran
ran on a synthesized input
junbongjang/contour-tracking/src/preprocessing/main_process_tracking_points.py
code served (permissive licence) · get_code("a92d51e83bd4b792")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned ndarray, ndim 2, dtype family float, shape (5, 2) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import numpy as np

# Create a sample contour points array
# This represents points of a contour with x,y coordinates
processed_ordered_contour_points = np.array([
    [10, 20],
    [15, 25],
    [20, 15],
    [25, 30],
    [12, 28]
], dtype=np.float32)

height = 100
width = 200

ARGS = {
    "processed_ordered_contour_points": processed_ordered_contour_points,
    "height": height,
    "width": width
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, ndim 2, dtype family float, shape (5, 2) — shape (5, 2) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
robust_l1 Ran
ran, honoured the contract we drafted
junbongjang/contour-tracking/src/tracking_utils.py
pointer only (licence: Apache-2.0); the stored copy get_code serves, from dangeng/CorrWiseLosses, is pointer only (licence: NONE) · get_code("94e47e87bb26b846")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (Apache-2.0), but get_code serves the stored copy, first harvested from dangeng/CorrWiseLosses, and GitHub detected no licence in that repository when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

robust_l2 Ran
ran on a synthesized input
junbongjang/contour-tracking/src/tracking_utils.py
code served (permissive licence) · get_code("8f9b339cc1ef3aa1")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned Tensor, ndim 2, dtype family float, shape (2, 4) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import numpy as np

torch.manual_seed(0)

# The function robust_l2 takes a single parameter x
# and performs: x ** 2 + 0.001 ** 2
# x should be a tensor that supports exponentiation

x = torch.randn(2, 4)

ARGS = {
    "x": x
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 2, dtype family float, shape (2, 4) — shape (2, 4) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
sample_few_tracking_points Ran
ran on a synthesized input
junbongjang/contour-tracking/src/preprocessing/MATLAB_tracking_points.py
code served (permissive licence) · get_code("6db48046c588fa46")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned ndarray, ndim 2, dtype family float, shape (10, 2) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import numpy as np

torch.manual_seed(0)

# contour_points should be a sequence-like object (list, array, or tensor)
# that supports slicing with [GLOBAL_left_offset:-GLOBAL_right_offset]
# The function expects it to work with indices, so a numpy array or torch tensor
# Let's use a numpy array with enough points (need > 20 points given offsets of 10 each)

contour_points = np.random.randn(30, 2)  # 30 points with 2D coordinates

NUM_SAMPLE_POINTS = 10

ARGS = {
    "contour_points": contour_points,
    "NUM_SAMPLE_POINTS": NUM_SAMPLE_POINTS
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, ndim 2, dtype family float, shape (10, 2) — shape (10, 2) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
spatial_accuracy Ran
ran on a synthesized input
junbongjang/contour-tracking/src/metrics.py
code served (permissive licence) · get_code("bd7b2699ab4e54af")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned float64, ndim 0, dtype family float; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import numpy as np

# Set random seed for reproducibility
np.random.seed(0)

# Create two sets of point coordinates (normalized or unnormalized, function will normalize)
# Using shape (5, 2) for 5 points with (x, y) coordinates
ps1 = np.array([
    [10.5, 20.3],
    [50.2, 60.1],
    [100.7, 150.5],
    [200.3, 250.2],
    [300.1, 350.9]
], dtype=np.float32)

ps2 = np.array([
    [11.2, 21.1],
    [51.0, 61.3],
    [99.8, 151.2],
    [201.5, 248.9],
    [299.5, 351.4]
], dtype=np.float32)

image_width = 640
image_height = 480
thresh = 0.05

ARGS = {
    "ps1": ps1,
    "ps2": ps2,
    "image_width": image_width,
    "image_height": image_height,
    "thresh": thresh
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type float64, ndim 0, dtype family float — no output shape is stated: observe_signature.py recorded float ndim 0, which disagrees with the fixture run
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
check_model_frozen Not yet run
unverified (harvested, no recorded run)
junbongjang/contour-tracking/src/contour_flow_main.py
code served (permissive licence) · get_code("36e04cd3ded65746")
How it ran

no invocation recorded.

Also present in other papers' code

Of 17 samples lifted from this paper's linked repositories, 17 carry a function identity and 0 do not. Of the 17 distinct functions with an identity, 1 are also present in other papers whose code we hold (0 with at least one paper dated earlier, 0 the same month, 1 later) and 16 are only here among the 52,977 papers whose code Syntology holds (as of 2026-09-26).

Only here means: among the 52,977 papers whose code Syntology holds (51,887 with lifted samples, 1,090 through their assemblies only), as of 2026-09-26, no other paper's lifted samples or current assembly bricks contain an identical function body after normalisation.

FunctionAlso present inEarliest-datedOthers named
robust_l1 1 other paper
1 later
2104.09498 (2021-04)
Comparing Correspondences: Video Prediction with Correspondence-wise Losses

Presence, not direction: an identical function body in two papers' repositories does not say who wrote it, which came from which, or that either paper's method rests on it. Matching is exact after normalisation, so a renamed or edited function is not matched. Dates are when papers appeared, not when code was written. Only .py files the harvester lifted are covered; notebooks are not read. A new-style arXiv id YYMM.NNNN dates a paper to the month 20YY-MM; otherwise Paper.year dates it to the year; otherwise it is undated. Dates are when the paper appeared, not when the code was written. Read 2026-11-22.

Repositories linked to this paper

Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.

Abstract

We systematically compare and analyze a set of key components in unsupervised optical flow to identify which photometric loss, occlusion handling, and smoothness regularization is most effective. Alongside this investigation we construct a number of novel improvements to unsupervised flow models, such as cost volume normalization, stopping the gradient at the occlusion mask, encouraging smoothness before upsampling the flow field, and continual self-supervision with image resizing. By combining the results of our investigation with our improved model components, we are able to present a new unsupervised flow technique that significantly outperforms the previous unsupervised state-of-the-art and performs on par with supervised FlowNet2 on the KITTI 2015 dataset, while also being significantly simpler than related approaches.

For agents

The same record, over MCP at https://syntology.ai/mcp:

get_harvested_code_for_paper("2006.04902")
get_code_for_paper("2006.04902")
have("2006.04902")

The run record, dated, one paper per request, free:

curl https://syntology.ai/api/ran/2006.04902.json

A badge for a README (the split and the date, never a ratio):

[![Syntology run record](https://syntology.ai/api/ran/2006.04902.svg)](https://syntology.ai/paper/2006.04902)

Connect an agent — have() is free.